Florence-2 vs InternVL3.5
Pick Florence-2 for small, English scene text and short labels where you need a box per text line, or when the same model must also do detection or captioning. Pick InternVL3.5 for document question answering and Markdown or JSON output, and when you want one model family you can size from 1B up to 241B-A28B. The main split is output: Florence-2 gives only text and boxes, while InternVL3.5 gives structured text but no boxes by default.
Florence-2microsoft/florence-2 | InternVL3.5opengvlab/internvl | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 0.8 s / page | 3.0 s / page |
| Best for | Lightweight OCR with boxes | Self-hosted document understanding |
| Inputs | images | images |
| Languages | English | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | ✓ Yes |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose Florence-2 if
- It is tiny at 0.23B (base) or 0.77B (large), so it suits lightweight edge OCR in English.
- OCR_WITH_REGION returns quadrilateral boxes per text line, which InternVL3.5 does not give by default.
- The same model also does captioning, detection and grounding, so one model can cover a mixed vision pipeline.
Choose InternVL3.5 if
- It can output plain text, Markdown or JSON, while Florence-2 returns only text and boxes with no tables or LaTeX.
- Nine sizes from 1B to 241B let you trade accuracy for cost, and models up to 30B fit on a single A100.
- It is a good fit for document question answering in a self-hosted multimodal pipeline.
Key differences
| Florence-2 | InternVL3.5 | |
|---|---|---|
| Licence | MIT | Apache-2.0, built on Apache-2.0 Qwen3 |
| Model size | 0.23B (base), 0.77B (large) | 1B to 241B-A28B; the 8B variant is 8.5B |
| How you run it | Self-hosted via Transformers; free, you pay for compute | Self-hosted via Transformers, LMDeploy or vLLM; free, you pay for compute |
| Output formats | Plain text and bounding boxes (quadrilateral boxes per text line) | Plain text, Markdown and JSON; no bounding boxes or layout output by default |
| Hardware | Not published | Up to 30B on a single A100; 38B needs two A100s; 241B-A28B needs eight A100s |
| Handwriting and non-English text | A Hugging Face blog test found spelling errors and missing spaces on cursive handwriting and French text; Chinese output is mostly wrong because the vocabulary lacks CJK characters | Not published |
| Published OCR benchmark | No OCR score listed; TextCaps CIDEr zero-shot (large) is 72.8 | OCR scores appear only as chart images on the model card, not as text values |
Sources
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
- Hugging Face model card (8B)
- InternVL GitHub
What each one is
Florence-2
Small Microsoft vision model (0.23B and 0.77B) with OCR and OCR-with-regions tasks that return text with boxes.
Florence-2 detailsInternVL3.5
Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.
InternVL3.5 detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "microsoft/florence-2","fallbacks": ["opengvlab/internvl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Florence-2 alternativesInternVL3.5 alternativesBest OCR models in 2026